The role of main sociodemographic variables in suicide ideation detection using machine learning

This study investigates the role of key sociodemographic variables in the detection of suicidal ideation within a Yucatan population, employing machine learning (ML) classification models. Data was obtained from the APPSI platform, encompassing psychological assessments (C-SSRS, DASS21, EAYIE) and sociodemographic information (age, gender, municipality, etc.). Ten classification algorithms were trained and evaluated to predict suicidal ideation. The Logistic Regression model demonstrated the strongest performance, achieving an F1 score of 0.77 and an accuracy of 81%, with a recall of 74% for the at-risk group. Results highlight the potential of ML to support mental health decision-making and the importance of sociodemographic factors in suicide ideation detection.